Using FSRS parameters to identify leech cards

I suppose I should have used suggestion tag but it’s related to FSRS too.

I have noticed recently that many of my leech cards(in my case that is 5 lapses) have a difficulty rating of 100. As far as I understand, higher difficulty=slower the rise of stability=more required effort to retain the info. This immediately suggests to me a potential change in the leech mechanism.
As the general practice in language learning communities go, leech cards are set to suspend. Perhaps with the previous SM2 algorithm the leech mechanism was somewhat inefficient in that 1. you have already spent a lot of time doing a certain card before it becomes leech 2. Funnily enough some of my leech cards actually feel easy to me and some of them don’t have very high difficulty ratings. My guess is some cards end up getting easy but the current system doesn’t know and doesn’t care.

I am certainly not very knowledgeable about fsrs so would like to hear everyone’s thoughts. I personally think (assuming that difficulty ratings decrease with more reviews) cards that maintain say, 98% to 100% difficulty for a certain period after should be tagged leech. It would ensure we are not wasting time on stuff to difficult and time-demanding. Well I still don’t have a very concrete idea in mind but that’s the gist of it.
@L.M.Sherlock Do you think FSRS parameters can be used for this thing?


Difficulty is a relative value in FSRS. For some users, most cards’ difficulty is 100%. Leech should has a more rigid threshold than difficulty.


Do you think it wouldn’t be possible to say, identify the cards taking the most time/effort with difficulty? That would require ranking the cards according to difficulty(?) and skimming off of some of it according to some standards. If a person has mostly 100 percent rated difficulty then ig it would be hard. Do you think maybe some other parameters can be used to identify leech?

For some cards, the 100% difficulty means the upper limitation of difficulty is 100. But the number of lapses doesn’t have a ceiling. So I recommend using the current leech detective method based on lapses.

1 Like

Maybe the Helper or Advanced Browser could search for “cards that maintain say, 98% to 100% difficulty for a certain period”, like d>=0.95:5r (>=95% in the last 5 repetitions). Although “:” is used with rated in a different sense.


There is a discussion of related methods at Suggestion: Lapse to Review Ratio

1 Like

It could also sort by the minimum difficulty over n most recent reviews (if the reviews have difficulty).

1 Like